When a visitor uses a website's search box, they do something valuable: they write, in their own words, what they are looking for. No other analytics data is this direct. Page views show where people went; internal searches show where they wanted to go. Yet on many sites this information is either not collected or sits in a report nobody opens.
What to record
Each search needs only a few pieces of data:
- the search term, normalised to lowercase and without extra spaces;
- the number of results returned;
- the page the search started from;
- what happened next: a click on a result, a new search, leaving the site.
If the internal search engine uses a URL parameter, such as «?q=», many analytics tools can read it directly. If search happens without a page change, you need a dedicated event with the term and number of results.
Zero-result searches
This is the first report to look at, and often the most useful. A search with no results is a question the site failed to answer. The causes usually fall into a few categories:
| Cause | Example | Action |
|---|---|---|
| Unrecognised synonym | «quote» instead of «cost estimate» | add synonyms to the engine |
| Typo | «configuartion» | typo tolerance |
| Missing content | a service that does not exist on the site | consider creating it |
| Plural or inflected term | «router» found, «home routers» not | improve text analysis |
| Code or product name | a specific model | index codes as well |
Every row in this table is a concrete action, and many take only a few minutes.
Refinements
When someone searches for one term and immediately searches for another, they are correcting their question. Pairs of consecutive searches tell a lot: if people searching «backup» then search «automatic backup», perhaps the results for «backup» show overly generic content. If people searching a service name then search «prices», information is missing from the service page.
Exits after search
A search followed by leaving the site is a sign of dissatisfaction, even when there were results. By comparing exit rates per term you find searches where results exist but are not the right ones, because they are badly ranked or because the most relevant content has an unclear title.
Search as an editorial guide
The most frequent queries, grouped by theme, are a priority list for whoever writes content. If a question recurs every week and the answer is buried at the bottom of a long page, that answer deserves its own page, or at least a section with a heading that uses the visitors' own words. It is the same principle as keyword research, with an advantage: these are words from people who are already on the site.
Search as a navigation signal
The starting page matters too. If many searches start from the home page and concern a service that already has a menu entry, that entry is not visible enough or uses a label different from what people expect. In this sense, site search analysis complements usability testing of the menu.
Privacy
People sometimes type personal data into the search box: an email address, a case number, their own name. Before sending the term to analytics, it is wise to filter anything that looks like an email, phone number or long string of digits and replace it with a placeholder. Keeping only the terms, without linking them to individual profiles, is enough for every analysis described here.
A monthly review
In my work I suggest a half-hour monthly review in this order:
- the twenty most frequent zero-result searches, with an action for each;
- the ten searches with the highest exit rate;
- new terms that appeared compared with the previous month;
- the pages that start the most searches.
Note every action: the following month you can check whether that search now produces clicks on results.
In short
- Site search shows, in visitors' own words, what they are looking for.
- Record the term, number of results, starting page and next action.
- Zero-result searches are the most useful report and the quickest to fix.
- Refinements and exits reveal results that exist but are not relevant enough.
- Filter personal data out of terms and review the reports every month.
